Re: BUG #5592: list of integer undefined behaviors

Tom, would you be willing to isolate these operations into functions
that could be marked with a "no_overflow_check" attribute? This would
be easy for us to deal with, would survive preprocesing cleanly, and
wouldn't have any performance cost since inliners do a fine job.

John

On 8/3/2010 3:43 PM, Tom Lane wrote:
> John Regehr<regehr(at)cs(dot)utah(dot)edu> writes:
>>> Just to follow up: all the other ones seem to be non-problems.
>
>> Would you folks be willing to specify which arithmetic operations are
>> considered to be safe in the case of overflow? Something simple like an
>> "INTEGER_OVERFLOW_OK" comment at the end of the line of code containing
>> the operation would suffice. This would let me automatically filter out
>> error messages on these lines of code in the future.
>
> That might be doable for individual operations, but I don't think that
> (for example) having to label all the users of RIGHTMOST_ONE() would be
> very maintainable. Is your code capable of tracking back to a macro
> definition?
>
> Also, it would be nicer if we could put the marker comment on an
> adjacent line. If it has to be on the same line then there are
> formatting problems when the code is wide (and pgindent could break it).
>
> regards, tom lane
>